ASK AI

ASK AI is a chat built into the dashboard (the AI tab) that answers questions about your workspace — runs, tests, machines and metrics — and renders answers as live dashboard components: tables, charts, stat cards and action buttons.

Architecture

• Bring your own model. The workspace owner or an admin configures an LLM endpoint under Settings → AI provider. Both OpenAI-compatible (/chat/completions) and Anthropic-compatible (Messages API) wire formats are supported. perfscale does not operate or proxy an LLM of its own.
• Encrypted credentials. The provider API key is encrypted at rest with AES-256-GCM and is never returned by any API — only its last 4 characters are shown.
• Your token, your data. The assistant fetches workspace data through the same REST API the dashboard uses, authenticated with the asking user's access token. RBAC and tenant isolation apply unchanged.
• MCP tools. The model's tools are the same composite MCP server that powers @perfscale/controlplane-mcp: reads (runs, tests, machines, metrics) plus writes — write_test, write_config, run_test — all gated by your RBAC permissions, so the assistant can only do what you can.
• Your MCP servers. Admins can attach remote MCP servers (Grafana, internal tools) under Settings → AI → MCP Servers; their tools join the chat for every member, prefixed with the server's name. Header secrets are encrypted at rest.

Generative UI

Answers are composed from a fixed component vocabulary — tables, line charts, stat cards, callouts — plus run actions: buttons that open the run page, jump to the metrics dashboard, or download the run's log file. The model supplies only data and the run ID; all URLs are constructed by the dashboard.

Threads

Conversations persist per user. Each thread has a stable URL:

/admin/ai/chat/<thread-id>

Stored chat history counts toward the workspace storage quota.

Plan limits

ASK AI is available on every plan. Monthly workspace data-tool calls are metered:

Plan	Tool calls / month
Starter	100
Growing	1,000
Enterprise	unlimited

When the budget is exhausted the chat returns a clear error; the counter resets each calendar month.

API

Endpoint	Auth	Purpose
GET /api/v1/ai/status	user	Whether a provider is configured (drives nav visibility)
GET/PUT/DELETE /api/v1/ai/settings	admin+	Provider configuration (key write-only)
GET /api/v1/ai/usage	user	Current month tool-call usage and limit
GET/POST/PUT/DELETE /api/v1/ai/threads*	user	Thread and message CRUD (per-user)